Won Hwa Kim
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Won Hwa Kim has a total of 159 co-authors. Here are the top co-authors by number of shared publications:
Guorong Wu
University of North Carolina at Chapel Hill
10 shared publications- Moo K. Chung
Department of Biostatistics and Medical Informatics, University of Wisconsin-Madison, USA
8 shared publications Nagesh Adluru
Waisman Laboratory for Brain Imaging and Behavior, University of Wisconsin - Madison, Madison, WI, USA
8 shared publicationsVikas Singh
Department of Biostatistics and Medical Informatics, University of Wisconsin - Madison, Madison, WI, USA; Department of Computer Science, University of Wisconsin - Madison, Madison, WI, USA
8 shared publications- Vikas Singh
University of Wisconsin-Madison, Madison, Wisconsin, USA
7 shared publications Sterling C. Johnson
6 shared publications
Latest Publications
- 27 September 2026
- Published by Springer Nature
- p. 525-535
Set-Inclusive Uncertainty Modeling for Robust Brain Tumor Segmentation
- 27 September 2026
- Published by Springer Nature
- p. 525-535
Set-Inclusive Uncertainty Modeling for Robust Brain Tumor Segmentation
- 26 September 2026
- Published by Springer Nature
- p. 181-191
ENC-ODE: Event-level Neurodegenerative Modeling in Continuous Time with Neural ODEs
- 26 September 2026
- Published by Springer Nature
- p. 181-191
ENC-ODE: Event-level Neurodegenerative Modeling in Continuous Time with Neural ODEs
- 26 September 2026
- Published by Springer Nature
- p. 557-567
SPHARM-Mamba: Rotation-Invariant Multiscale Modeling for Brain Age Prediction
- 26 September 2026
- Published by Springer Nature
- p. 557-567
SPHARM-Mamba: Rotation-Invariant Multiscale Modeling for Brain Age Prediction
- 26 September 2026
- Published by Springer Nature
- p. 546-556
Spectral-Temporal State Space Modeling on Functional Brain Networks
- 26 September 2026
- Published by Springer Nature
- p. 546-556
Spectral-Temporal State Space Modeling on Functional Brain Networks
- 25 September 2026
- Published by Springer Nature
- p. 563-572
SurgMark: Structure-Aware Watermarking for 4D Gaussian Splatting in Surgical Scenes
- 25 September 2026
- Published by Springer Nature
- p. 563-572
SurgMark: Structure-Aware Watermarking for 4D Gaussian Splatting in Surgical Scenes
- 25 September 2026
- Published by Springer Nature
- p. 552-562
Multi-frame Restoration for 10 Hz Lissajous Confocal Laser Endomicroscopy
- 25 September 2026
- Published by Springer Nature
- p. 552-562
Multi-frame Restoration for 10 Hz Lissajous Confocal Laser Endomicroscopy
- 25 September 2026
- Published by Springer Nature
- p. 274-284
Learning Robust Medical Image Segmentation Under Mixed-Quality Annotations
- 25 September 2026
- Published by Springer Nature
- p. 274-284
Learning Robust Medical Image Segmentation Under Mixed-Quality Annotations
- 11 September 2026
- Published by Springer Nature
- p. 76-93
Residual-Guided Expert Specialization for Incomplete Multimodal Learning
- 11 September 2026
- Published by Springer Nature
- p. 76-93
Residual-Guided Expert Specialization for Incomplete Multimodal Learning
- 22 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. PP (01628828) , 1-12
A Variational Mean-Field Control Framework for Graph Representation Learning
Feature representation learning in graph neural networks (GNNs) is a dynamic process driven by progressive information exchange throughout the graph. Current GNNs typically apply pre-defined message-passing heuristics uniformly across all graph data, even when the assumed relational inductive bias conflicts with the intrinsic graph properties (e.g., heterophily). This limitation hinders a principled understanding of graph learning and motivates a shift from application-specific architectures toward a generalizable paradigm. In this work, we conceptualize GNN learning through the lens of mean-field control, where the desired learning outcome corresponds to a variational critical point, each node reaches a representation that is adaptively calibrated to both itself and the entire graph. Building on this formulation, we propose a mathematical framework based on mean-field control (MFC) to adaptively design GNNs for unseen graph data. Specifically, we jointly optimize two control patterns: diffusive mobility, governing information propagation across the graph, and reactive mobility, regulating feature transformation at individual nodes. Both controls are learned from each input instance by solving an MFC variational problem through Hamiltonian flows characterized by partial differential equations (PDE). Our variational framework unifies existing PDE-based GNNs as special mean-field games with fixed control patterns, and yields an end-to-end deep model, termed Nash-GNN. Extensive experiments show that Nash-GNN achieves state-of-the-art performance across diverse benchmarks, including heterophilic graphs and human connectomes. Beyond empirical gains, the MFC framework offers a principled lens to examine the mechanisms of graph representation learning through collective dynamics, opening promising directions for next-generation GNN design.
- 22 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. PP (01628828) , 1-12
A Variational Mean-Field Control Framework for Graph Representation Learning
Feature representation learning in graph neural networks (GNNs) is a dynamic process driven by progressive information exchange throughout the graph. Current GNNs typically apply pre-defined message-passing heuristics uniformly across all graph data, even when the assumed relational inductive bias conflicts with the intrinsic graph properties (e.g., heterophily). This limitation hinders a principled understanding of graph learning and motivates a shift from application-specific architectures toward a generalizable paradigm. In this work, we conceptualize GNN learning through the lens of mean-field control, where the desired learning outcome corresponds to a variational critical point, each node reaches a representation that is adaptively calibrated to both itself and the entire graph. Building on this formulation, we propose a mathematical framework based on mean-field control (MFC) to adaptively design GNNs for unseen graph data. Specifically, we jointly optimize two control patterns: diffusive mobility, governing information propagation across the graph, and reactive mobility, regulating feature transformation at individual nodes. Both controls are learned from each input instance by solving an MFC variational problem through Hamiltonian flows characterized by partial differential equations (PDE). Our variational framework unifies existing PDE-based GNNs as special mean-field games with fixed control patterns, and yields an end-to-end deep model, termed Nash-GNN. Extensive experiments show that Nash-GNN achieves state-of-the-art performance across diverse benchmarks, including heterophilic graphs and human connectomes. Beyond empirical gains, the MFC framework offers a principled lens to examine the mechanisms of graph representation learning through collective dynamics, opening promising directions for next-generation GNN design.
- Open Access
- 7 May 2026
- Published by MDPI AG in Diagnostics
- Vol. 16 (10) , 1422
A Transformer-Based Machine Learning Framework for Risk Stratification of Left Bundle Branch Block After Transcatheter Aortic Valve Replacement
Background/Objectives: Left bundle branch block (LBBB) remains a common complication after transcatheter aortic valve replacement (TAVR) and is associated with adverse clinical outcomes. However, accurate prediction of LBBB remains challenging due to the complex interactions among the anatomical, procedural, and clinical factors. This study aimed to develop a machine learning (ML)-based framework to predict LBBB and identify relevant contributing features. Methods: In this multicenter retrospective study, we analyzed 242 patients undergoing TAVR across three institutions. A machine learning framework incorporating transformer-based feature selection and conventional classifiers was developed. Model performance was evaluated using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC). Internal validation was performed using bootstrap resampling. Results: The gradient boosting model using ML-derived features demonstrated the most balanced performance, achieving an accuracy of 78.05% and an F1-score of 50.46%, with modest discrimination (AUC 0.61). The ML-based approach identified clinically relevant features, including coronary height, left ventricular outflow tract/annulus ratio, and prosthetic valve size, as well as additional variables not emphasized in conventional analyses. Conclusions: ML-based feature selection can capture complex feature interactions beyond traditional statistical approaches and provide clinically meaningful insights into risk stratification for LBBB after TAVR. Although predictive performance was modest, this approach highlights the potential of ML for improved risk stratification and individualized procedural planning. Further large-scale external validation is warranted.
- 7 May 2026
- Published by MDPI AG in Diagnostics
- Vol. 16 (10) , 1422
Open AccessA Transformer-Based Machine Learning Framework for Risk Stratification of Left Bundle Branch Block After Transcatheter Aortic Valve Replacement
Background/Objectives: Left bundle branch block (LBBB) remains a common complication after transcatheter aortic valve replacement (TAVR) and is associated with adverse clinical outcomes. However, accurate prediction of LBBB remains challenging due to the complex interactions among the anatomical, procedural, and clinical factors. This study aimed to develop a machine learning (ML)-based framework to predict LBBB and identify relevant contributing features. Methods: In this multicenter retrospective study, we analyzed 242 patients undergoing TAVR across three institutions. A machine learning framework incorporating transformer-based feature selection and conventional classifiers was developed. Model performance was evaluated using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC). Internal validation was performed using bootstrap resampling. Results: The gradient boosting model using ML-derived features demonstrated the most balanced performance, achieving an accuracy of 78.05% and an F1-score of 50.46%, with modest discrimination (AUC 0.61). The ML-based approach identified clinically relevant features, including coronary height, left ventricular outflow tract/annulus ratio, and prosthetic valve size, as well as additional variables not emphasized in conventional analyses. Conclusions: ML-based feature selection can capture complex feature interactions beyond traditional statistical approaches and provide clinically meaningful insights into risk stratification for LBBB after TAVR. Although predictive performance was modest, this approach highlights the potential of ML for improved risk stratification and individualized procedural planning. Further large-scale external validation is warranted.